Inside Claude’s J-Space: Digital Minds Reading Group

Join DC Digital Minds for the first meeting of our new monthly reading group, where we’ll gather to explore important new work on AI minds, behavior, cognition, and moral status.

For our inaugural discussion, we’ll discuss Anthropic’s blog post, “A Global Workspace in Language Models.” The researchers identify a collection of internal representations they call the J-space, which appears to make certain information available across different parts of the model for reporting, reasoning, and guiding behavior. Anthropic connects these findings to the idea of access consciousness, a functional concept drawn from philosophy and cognitive science.

What does that mean? How strong is the evidence? What does the paper establish, and what remains uncertain? We'll carefully digest what Anthropic has put forward, understand the methods and claims, ask good questions, and decide for ourselves what we think.

Beginners Welcome
No specialized technical background is required. Whether you work professionally in AI, approach these questions through philosophy or social science, or are simply find it fascinating, you're warmly invited. We’ll work through the paper together rather than expecting everyone to arrive as an expert.
Come for the interpretability research -- stay for the interesting people and thoughtful conversation.

Before the Meeting
Please read Anthropic’s blog post, linked above. Those who would like to go deeper are encouraged, but not required, to read the full paper, "Verbalizable Representations Form a Global Workspace in Language Models."

This will be the first gathering of a monthly DC Digital Minds reading group, on the second Thursday of the month. Future selections will explore research and ideas relevant to understanding increasingly complex AI systems and navigating their place in human society.